Extraction and Classification of Human Gait Features

download Extraction and Classification of Human Gait Features

of 11

Transcript of Extraction and Classification of Human Gait Features

  • 8/9/2019 Extraction and Classification of Human Gait Features

    1/11

     

    H. Badioze Zaman et al. (Eds.): IVIC 2009, LNCS 5857, pp. 596–606, 2009.

    © Springer-Verlag Berlin Heidelberg 2009

    Extraction and Classification of Human Gait Features

    Hu Ng1, Wooi-Haw Tan2, Hau-Lee Tong1, Junaidi Abdullah1,

    and Ryoichi Komiya 3 

    1 Faculty of Information Technology, Multimedia University, Persiaran Multimedia,

    63100 Cyberjaya, Selangor Malaysia2 Faculty of Engineering, Multimedia University, Persiaran Multimedia, 63100 Cyberjaya,

    Selangor Malaysia3 Department of Mechatronic and BioMedical Engineering, Faculty of Engineering and Science,

    Universiti Tunku Abdul Rahman, Jalan Genting Kelang, Setapak 53300 Kuala Lumpur{nghu,twhaw,hltong,junaidi}@mmu.edu.my, [email protected] 

    Abstract. In this paper, a new approach is proposed for extracting human gait

    features from a walking human based on the silhouette images. The approach

    consists of six stages: clearing the background noise of image by morphological

    opening; measuring of the width and height of the human silhouette; dividing

    the enhanced human silhouette into six body segments based on anatomical

    knowledge; applying morphological skeleton to obtain the body skeleton; ap-

    plying Hough transform to obtain the joint angles from the body segment skele-

    tons; and measuring the distance between the bottom of right leg and left leg

    from the body segment skeletons. The angles of joints, step-size together withthe height and width of the human silhouette are collected and used for gait

    analysis. The experimental results have demonstrated that the proposed system

    is feasible and achieved satisfactory results.

    Keywords: Human identification, Gait analysis, Fuzzy k-nearest neighbour.

    1 Introduction

    Personal identification or verification schemes are widely used in systems that require

    determination of the identity of an individual before granting the permission to access

    or use the services. Human identification based on biometrics refers to the automatic

    recognition of the individuals based on their physical and/or behavioural characteris-

    tics such as face, fingerprint, gait and spoken voice. Biometrics are getting important

    and widely acceptable nowadays because they are really personal / unique that one

    will not lose or forget it over time.

    Gait is unique, as every individual has his/her own walking pattern. Human walk-

    ing is a complex locomotive action which involves synchronized motions of bodyparts, joints and the interaction among them [1]. Gait is a new motion based biometric

    technology, which offers the ability to identify people at a distance when other bio-

    metrics are obscured. Furthermore, there is no point of contact with any feature cap-

    turing device and is henceforth unobtrusive.

  • 8/9/2019 Extraction and Classification of Human Gait Features

    2/11

      Extraction and Classification of Human Gait Features 597

    Basically, gait analysis can be divided into two major categories, namely model-

    based method and model-free method. Model-based method generally models the

    human body structure or motion and extracts features to match them to the model

    components. The extraction process involves a combination of information on the

    human shape and dynamics of human gait. This implies that the gait dynamics areextracted directly by determining joint positions from model components, rather than

    inferring dynamics from other measures, thus, reducing the effect of background

    noise (such as movement of other objects). For instance, Johnson used activity-

    specific static body parameters for gait recognition without directly analyzing gait

    dynamics [2]. Cunado used thigh joint trajectories as the gait features [3]. The advan-

    tages of this method are the ability to derive gait signatures directly from model pa-

    rameters and free from the effect of different clothing or viewpoint. However, it is

    time consuming and the computational cost is high due to the complex matching and

    searching process.On the other hand, model-free method normally distinguishes the entire human

    body motion using a concise representation without considering the underlying struc-

    ture. The advantages of this method are low computational cost and less time con-

    suming. For instance, BenAbdelkader proposed an eigengait method using image self-

    similarity plots [4]. Collins established a method based on template matching of body

    silhouettes in key frames during a human’s walking cycle [5]. Philips characterized

    the spatial-temporal distribution generated by gait motion in its continuum [6].

    This paper presents the unique concept of extracting the human gait features of

    walking human from consecutive silhouette images. First, the height and width of thehuman subject are determined. Next, each human silhouette image is enhanced and

    divided into six body segments to construct the two-dimension (2D) skeleton of the

    body model. Then, Hough transform technique is applied to obtain the joint angle for

    each body segment. The distance between the bottoms of both lower legs can also be

    obtained from the body segment skeletons. This concept of joint angle calculation is

    found faster in process and less complicated than the model-based method like linear

    regression approach by Yoo [7] and temporal accumulation approach by Wagg [8].

    2 Overview of the System

    First, morphological opening is applied to reduce background noise on the raw human

    silhouette images. The width and height of each human silhouette are then measured.

    Next, each of the enhanced human silhouettes is divided into six body segments based

    on the anatomical knowledge [10]. Morphological skeleton is later applied to obtain

    the skeleton of each body segment. The joint angles are obtained after applying

    Hough transform on the skeletons. Step-size, which is the distance between the bot-

    toms of both legs are also measured from the skeletons of the lower legs. The dimen-

    sion of the human silhouette, step-size and six joint angles from body segments –

    head and neck, torso, right hip and thigh, right lower leg, left hip and thigh, and left

    lower leg are then used as the gait features for classification. Fig. 1 summarizes the

    process flow of the proposed system.

  • 8/9/2019 Extraction and Classification of Human Gait Features

    3/11

    598 H. Ng et al.

    Fig. 1. Flow chart of the proposed system

    2.1 Original Image Enhancement

    The acquired original raw human silhouette images are obtained from the small sub- ject gait database, University of Southampton [9]. They used static cameras to cap-

    ture eleven subjects walking along the indoor track in four different angles. Videodata was first preprocessed using Gaussian averaging filter for noise suppression,followed by Sobel edge detection and background subtraction technique to create thehuman silhouette images.

    Due to poor lighting condition during the video shooting, shadow was foundespecially near to the feet. It appeared as part of the subject body in the binary humansilhouette image as shown in Fig. 2. The present of the artefact affects the gait featureextraction and the measurement of joint angles. This problem can be reduced by applying

    morphological opening with a 7×7 diamond shape structuring element, as denoted by

    AB = (AB) ⊕ B) . (1)

    where A is the image and B is the structuring element. The opening first performs

    erosion operation and followed by dilation operation. Fig. 2 shows the result of apply-

    ing morphological opening on a human silhouette image.

    (a) Original image (b) Enhanced image

    Fig. 2. Original and enhanced image after morphological opening

    Measurement

    of width and

    height

    Measurement

    of step-sizeSkeletonization

    of body

    segments

    Human

    silhouette

    segmentation

    Joint angles

    extraction

    Original

    image

    enhancement

    Computationof the

    similarities

    Determinationof the k-nearest

    neighbor

    Classificationof the unlabeled

    subjects

  • 8/9/2019 Extraction and Classification of Human Gait Features

    4/11

      Extraction and Classification of Human Gait Features 599

    2.2 Measurement of Width and Height

    The width and height of the subject during the walking sequences are measured from

    the bounding box of the enhanced human silhouette, as shown in Fig. 3. These two

    features will be used for gait analysis in the later stage.

    Height

    Width  

    Fig. 3. The width and height of a human silhouette

    2.3 Dividing Human Silhouette

    At this stage, the enhanced human silhouette is divided into six body segments based

    on the anatomical knowledge [10]. First, the centroid of the subject is determined by

    calculating the centre of mass of the human silhouette. The area above the centroid is

    considered as the upper body – head, neck and torso. The area below the centroid is

    considered as the lower body – hips, legs and feet.

    Next, one third of the upper body is divided as the head and neck. The remaining

    two thirds of the upper body are classified as the torso. The lower body is divided into

    two portions – (i) hips and thighs (ii) lower legs and feet with the ratio one to two.

    Again, the centroid coordinate is used to divide the two portions into the final four

    segments – (i) right hip and thigh (ii) lower right leg and foot (iii) left hip and thigh

    and (iv) lower left leg and foot.

    Fig. 4. Six body segments

  • 8/9/2019 Extraction and Classification of Human Gait Features

    5/11

    600 H. Ng et al.

    Fig. 4 shows the six segments of the body, where “a” represent head and neck, “b”represents torso, “c” represents right hip and thigh, “d” represents lower right leg andfoot, “e” represents left hip and thigh and “f” represents lower left leg and foot.

    2.4 Skeletonization of Body Segments

    To better represent each body segment, morphological skeleton is used to constructthe skeleton for each of the body segments. Skeletonization involves consecutiveerosions and opening operations on the image until the set difference between the twooperations is zero. 

    Erosion Opening Set differences

    Ak B (Ak B) B (Ak B) – ((Ak B)) B (2)

    where A is an image, B is the structuring element and k  is from zero to infinity. Fig. 5.shows the skeleton of the body segments.

    Fig. 5. Skeleton on a torso segment

    2.5 Joint Angles Extraction

    To extract the joint angle for each body segment, Hough transform is applied on the

    skeleton. Hough transforms maps pixels in the image space to the straight line

    Fig. 6. Joint angle formation

  • 8/9/2019 Extraction and Classification of Human Gait Features

    6/11

      Extraction and Classification of Human Gait Features 601

    through a parameter space. The skeleton in each body segment, which is the longest

    line, is indicated by the highest intensity point in the parameter space. Fig. 6 shows

    the joint angle formation from the most probable straight line detected via Hough

    transform, where φ  is the joint angle calculated using

    φ θ   =+°90  . (3)

    2.6 Measurement of Step-Size

    To obtain the step-size of each walking sequence, the Euclidian distance between the

    bottom ends of lower right leg and lower left leg are measured.

    Fig. 7 shows all the gait features extracted from a human silhouette, where Angle 7

    is the thigh angle, calculated as

    Angle 7 = Angle 6 – Angle 4 . (4)

     

    Fig. 7. All the extracted gait features

    3 Classification Technique

    For the classification, the supervised fuzzy K-Nearest Neighbour (KNN) algorithm is

    applied, as there is sufficient data to be used for training and testing. Basically, KNN

    is a classifier to distinguish the different subjects based on the nearest training data in

    the feature space. In other words, subjects are classified according to the majority of

    nearest neighbours.

  • 8/9/2019 Extraction and Classification of Human Gait Features

    7/11

    602 H. Ng et al.

    In extension to KNN, J.M. Keller [11] has integrated the fuzzy relation with the

    KNN. According to the Keller’s concept, the unlabeled subject’s membership func-

    tion of class i is given by Equation (5).

    2

    1

    2

    1

    1( )

    || ||( )

    1

    || ||

    ∈−

    ∈−

    ⎛ ⎞⎜ ⎟⎜ ⎟⎜ ⎟

    −⎝ ⎠=

    ⎛ ⎞⎜ ⎟⎜ ⎟⎜ ⎟

    −⎝ ⎠

    i

     x KNN  m

    i

     x KNN  m

    u x 

     x x u x 

     x x 

     . (5)

    where  x  ,  x   and U i( x ) represent the unlabelled subjects,  labelled subjects and  x ’s

    membership of class i respectively. Equation (5) will compute the membership value

    of unlabeled subject by the membership value of labelled subject and distance be-

    tween the unlabelled subject and KNN labelled subjects.

    Through the fuzziness, the KNN will annotate the appropriate class to the unla-

    belled subject by sum of similarities between labelled subjects. The algorithm in-

    volved in the identification of the human beings is implemented as follows:

    Step 1: Compute the distance between the unlabelled subject, and all labelled  or

    training subjects. The distance between an unlabelled subject x i and labelled

    subject, x  j is defined as:

    D ( x i, x  j) = || x i, x  j || − 2 . (6)

    Step 2: Sort the objects based on the similarity and identify the k-nearest neighbours.

    k-nearest neighbours, KNN ={ x 1, x 2, …, x k } . (7)

    Step 3: Compute the membership value for every class using Equation (5).

    Step 4: Classify unlabelled subject to the class with the maximum membership value

    as shown in Fig. 8.

    Fig. 8. An example for four nearest neighbours

  • 8/9/2019 Extraction and Classification of Human Gait Features

    8/11

      Extraction and Classification of Human Gait Features 603

    From Fig. 8, the values on the lines denote the similarities between unlabelled and

    labelled subjects. The sum of membership values for Class 1, m1=0.7 and for Class 2,

    m2=0.3. Since m1 is more than m2, so the unlabelled subject is classified as Class 1.

    4 Experimental Results and Discussion

    The experiment was carried out for nine subjects with three different conditions,

    which are walking with normal speed, walking in own shoes and walking in boots.

    The major objective was to identify the degree of accuracy for fuzzy KNN technique  

    by using the different values of k. For each subject, there were approximately twenty

    sets of walking data in normal track (walking parallel to the static camera). 

    In order to obtain the optimized results, five features were adopted for the classifi-

    cation. First, maximum thigh angle, θ max

      was determined from all the thigh anglescollected during a walking sequence. When θ max was located, the corresponding val-

    ues for the step-size, S  and width, w  and height, h are determined as well. From the

    graph plotted, it can be observed that the width for each subject changes in a sinusoi-

    dal patter over time, as shown in Fig. 9. Therefore, the last employed feature is the

    average of the maximum width, AP.

    Fig. 9. Graph of width versus time

    All the features were channelled into the classification process and the distance of

    similarity between unlabelled object,  x i and labelled object,  x  j was defined by Equa-

    tion (8).

    max max 2 2 2 2 2( , ) ( ) ( ) ( ) ( ) ( ) p pi j i j i j i j i j i j

     D x x w w h h S S A Aθ θ = − + − + − + − + −  . (8)

    The adopted algorithm was supervised fuzzy KNN, which requires training and test-ing. For the training part, a minimum of eight set of walking data were used for each

    subject. The residual data were used for the testing. The allocation of the training and

    testing data for each condition is shown in Table 1.

  • 8/9/2019 Extraction and Classification of Human Gait Features

    9/11

    604 H. Ng et al.

    Table 1. Allocation of the data for each condition

    Testing data set Training data set

    Normal speed 106 78

    Wearing own shoes 101 79

    Wearing boots 100 74

    Different values of k nearest neighbours were adopted for the classification, where

    k = 3, 4, 5, 6, 7 and 8. Since the minimum of the training data is eight, the maximum

    value of k was set to eight. The results obtained are depicted in Fig. 10 and Table 2.

    Fig. 10. Graph for the percentage of accuracy versus the value of k

    Table 2. The percentage of accuracy for fuzzy KNN

    k Normal speed

    (%)

    Wearing own shoes

    (%)

    Wearing boots

    (%)

    3 78.3 72.3 834 76.4 75.2 82

    5 75.5 72.3 81

    6 75.5 71.3 82

    7 76.4 71.3 81

    8 77.4 72.3 80

    From Table 2, it can be concluded that the changes of value k do not have a sig-

    nificant impact on the accuracy of the classification. However, when k = 3, results

    were slightly better on others. More satisfactory classification results might be ob-tained if more features are employed.

    In addition to evaluation for each condition, classification results for each subject

    were evaluated as well. This was to determine which unlabelled subjects were well

    identified and vice versa for all the conditions. Since k = 3 provided the best result for

    all three conditions, the experiment was carried out using k = 3 for subject evaluation.

  • 8/9/2019 Extraction and Classification of Human Gait Features

    10/11

      Extraction and Classification of Human Gait Features 605

    The obtained results for respective subject are shown in Table 3. From Table 3, sub-

     ject 1 produces the most satisfactory classification results for all three conditions. This

    was contributed by the adopted features for subject 1 which was highly distinctive

    from other subjects. Furthermore, there were not many variations between training

    and testing data for subject 1. In other words, subject 1 was well recognizable underthese three conditions. For the rest of the subjects, the accuracy of the classification

    was highly depending on the conditions. For instance, under normal speed the accu-

    racy for subject 8 only achieved an accuracy of 36.4%. This was due to the large

    number of misclassifications of subject 8 to other subjects.

    Table 3. The percentage of the classification results for three conditions when k = 3

    Normal speed

    (%)

    Wearing own shoes

    (%)

    Wearing boots

    (%)

    Subject 1 100 81.8 100Subject 2 81.8 58.3 90.9

    Subject 3 92.3 70 81.8

    Subject 4 80 37.5 81.8

    Subject 5 100 53.8 90.9

    Subject 6 100 90.9 45.5

    Subject 7 57.1 85.7 85.7

    Subject 8 36.4 80 91.7

    Subject 9 50 83.3 90

    5 Conclusion

    We have described a new approach for extracting the gait features from enhanced

    human silhouette image. The gait features is extracted from human silhouette by de-

    termining the skeleton from body segment. The joint angles are obtained after apply-

    ing Hough transform on the skeleton. In the future, more gait features will be ex-

    tracted and applied in order to achieve higher accuracy of classification.

    Acknowledgment

    The authors would like to thank Prof Mark Nixon, School of Electronics and Com-

    puter Science, University of Southampton, United Kingdoms for providing the data-

    base for use in this work.

    References

    1. 

    BenAbdelkader, C., Culter, R., Nanda, H., Davis, L.: EigenGait: Motion-based Recogni-tion of People Using Image Self-similarity. In: Proceeding of International Conference

    Audio and Video-Based Person Authentication, pp. 284–294 (2001)

    2. 

    Bobick, A., Johnson, A.: Gait Recognition Using Static, Activity-specific Parameters. In:

    Proceeding IEEE Computer Vision and Pattern Recognition, pp. 423–430 (2001)

  • 8/9/2019 Extraction and Classification of Human Gait Features

    11/11

    606 H. Ng et al.

    3. 

    Cunado, D., Nixon, M., Carter, J.: Automatic Extraction and Description of Human Gait

    Models for Recognition Purposes. Computer and Vision Image Understanding 90, 1–41

    (2003)

    4.  BenAbdelkader, C., Cutler, R., Davis, L.: Motion-based Recognition of People in Eigen-

    gait Space. In: Proceedings of Fifth IEEE International Conference, pp. 267–272 (2002)5.

     

    Collin, R., Gross, R., Shi, J.: Silhouette-based Human Identification from Body Shape and

    Gait. In: Proceedings of Fifth IEEE International Conference, pp. 366–371 (2002)

    6.  Phillips, P.J., Sarkar, S., Robledo, I., Grother, P., Bowyer, K.: The Gait Identification

    Challenge Problem: Dataset and Baseline Algorithm. In: Proceedings of 16th International

    Conference Pattern Recognition, pp. 385–389 (2002)

    7.  Yoo, J.H., Nixon, M.S., Harris, C.J.: Extracting Human Gait Signatures by Body Segment

    Properties. In: Fifth IEEE Southwest Symposium on Image Analysis and Interpretation,

    pp. 35–39 (2002)

    8.  Wagg, D.K., Nixon, M.S.: On Automated Model-based Extraction and Analysis of Gait.

    In: Proceedings of 6th IEEE International Conference on Automatic face and Gesture Rec-ognition, pp. 11–16 (2004)

    9.  Shutler, J.D., Grant, M.G., Nixon, M.S., Carter, J.N.: On a Large Sequence-based Human

    Gait Database. In: Proceedings of 4th International Conference on Recent Advances in Soft

    Computing, pp. 66–71 (2002)

    10. 

    Dempster, W.T., Gaughran, G.R.L.: Properties of Body Segments Based on Size and

    Weight. American Journal of Anatomy 120, 33–54 (1967)

    11. 

    Keller, J., Gray, M., Givens, J.: A Fuzzy K-Nearest Neighbour Algorithm. IEEE Trans.

    Systems, Man, Cybern. 15, 580–585 (1985)